Detecting Pedestrian in Clutter Scene Using Shape and Motion

This paper describes a pedestrian detection framework that is capable of achieving high detection rate and low false positive rate, especially in a video surveillance system with a clutter background. On the premise that the targets are active in the scene, motion cues are exploited to exclude the background positions from HOG evaluation. Since clutter background may contain gradient information similar to that of human body, this step can significantly reduce the false positive rate. The detection window is partitioned into four parts. The local features of each part are encoded using the HOG descriptors, and K-means algorithm is employed to learn the clusters for each part. Adaboost is used to select features and train the weak classifiers. A cascade structure strong classifier is then constructed to perform the detection. A set of experiments demonstrate the efficiency and performance of the proposed pedestrian detection system.